README.md

metadata

license: cc0-1.0
task_categories:
  - text-classification
  - question-answering
language:
  - en
tags:
  - logistics
  - supply-chain
  - benchmark
  - transportation

LogistikaBench Dataset Structure 🚚

LogistikaBench is a specialized evaluation benchmark designed to test the domain knowledge and reasoning capabilities of AI models in logistics, transport, and supply chain management. The dataset contains 1,446 multiple-choice questions across supply chain domains:

• Material Handling
• Transport
• Warehousing
• Supply Chain Management
• Procurement

More than one right answer can be available. For instance, [0, 2] in the answer column means the first and the third answers are correct.

No questions or answers in this dataset were generated using AI (such as large language models). The questions are mostly based on the following open-access textbooks and other free sources:

1. Caplice, C., & Ponce, E. (2023). MITx MicroMasters Program in SCM Key Concepts. MIT Center for Transportation & Logistics.
2. Ivanov, D., Tsipoulanidis, A., & Schönberger, J. (2017). Global supply chain and operations management. A decision-oriented introduction to the creation of value. ( http://global-supply-chain-management.de/Slides)
3. Koningsveld, M., Verheij, H. J., Taneja, P., & de Vriend, H. J. (2021). Ports and Waterways: Navigating the changing world.
4. Carpenter, M. A., & Dunung, S. P. (2012). Challenges and opportunities in international business. Creative Commons by-nc-sa, https://2012books.lardbucket.org/
5. EU/UN/ITF/OECD. (2019). Glossary for Transport Statistics.

Limitations & Potential Errors

All test items are derived strictly from curated, authentic source materials used in actual university-level examinations and assessments. Every question reflects pedagogical evaluations designed for real academic environments.

While rigorous curation was applied, human and structural limitations are inevitable:

Answers: Human error during transcription or answer key compilation means  inaccuracies are possible and even likely.
Subject Categorization: Due to the  human error or multidisciplinary nature of the domain with  overlapping category.

Call for Community Improvement

To maintain dataset quality, we encourage the community, educators, and researchers to help identify and correct these flaws. You can contribute improvements in the following standard ways:

GitHub Pull Requests: Submit direct corrections for answer keys or categorization via  official repository.

Dataset Discussions / Issues: Report ambiguous questions, potential answer errors, or classification misalignments on the Kaggle or Hugging Face community tabs.

🚀 Quick Start: Run the Benchmark in Google Colab

Evaluate an open-source model on LogistikaBench for free using a Google Colab T4 GPU

1. Click the button below to open a fresh notebook:
2. Copy the evaluation script evaluate.py` file.
3. Paste it into your Colab cell, change the model_id to model you want to test, and hit Play!

🤝 How to Submit Scores (Leaderboard Contributions)

Add scores for proprietary models (e.g., GPT, Claude, Grok, Gemini) or your own fine-tuned models - contribute to the leaderboard by submitting a Pull Request:

1. Run the evaluation script on your end using your own API keys.
2. Edit the README.md file directly on this dataset repository.
3. Add your model details to the Leaderboard markdown table in alphabetical or ranked order:
   • Model Name | Provider | Strict Match Accuracy (%) | Contributor / Handle
4. Open a Pull Request. Once reviewed and merged, your score will appear live on the official leaderboard

---